Cheilospirura hamulosa in the Rock Partridge (Alectoris graeca saxatilis): Epidemiological Patterns and Prediction of Parasite Distribution in France

The rock partridge (Alectoris graeca saxatilis) is an alpine Galliform with high conservation value. Several factors, including parasitic helminths, play a role in population dynamics, and consequently in the conservation management of wild Galliformes. The aim of this study was to assess the epidemiological characteristics of Cheilospirura hamulosa (Nematoda, Acuarioidea) in the Rock partridge population in France. Machine learning modeling algorithms were applied to identify the environmental variables influencing parasite occurrence, and to map parasite presence probability. The present work is based on a long-term sampling (1987–2019) conducted in the French Alps. C. hamulosa was found with a prevalence (P) of 39% (Confidence Interval—CI 95%: 34–43), and mean intensity of 7.7 (7.8 sd). The highest prevalence (P: 67%, CI 95%: 54–80) was detected in the period 2005–2009. Latitude was the most important variable shaping the parasite distribution, followed by altitude, annual mean temperature, temperature seasonality, and the amount of precipitation of the coldest quarter. The area suitable for parasite presence included 73% of the French Alps. This work represents the first epidemiological surveillance on C. hamulosa infection in the rock partridge. It provides evidence of a high level of infection and identifies priority areas at higher infection risk, where a close monitoring of the rock partridge populations should be carried out.


Introduction
The rock partridge (Alectoris graeca saxatilis) is an alpine Galliform inhabiting the prairies and rocky meadows from the French Alps to Slovenia [1]. Like other alpine Galliformes such as the rock ptarmigan (Lagopus muta), it has recently suffered a significant decline in the Alps [2], (and it is now listed in Annex I of the "Bird" Directive 2009/147/EC, aiming to preserve species and their habitats. The species A. graeca is classified as "Near threatened" (NT) in the last update of the IUCN Red list, in view of the fact that the population is significantly declining, particularly in the Eastern part of its range [3]. Most of its global population have suffered a reduction exceeding 30% of the estimated consistency over the last three generations [4]. The same classification (NT) appeared in the last update of the French National Red book [5].
The species' long-term conservation is negatively influenced by several factors, including habitat degradation [6], reduction of habitat suitability due to changes in traditional agro-pastoral activities [7], and fragmentation of habitat, with decreased connection among metapopulations [8]. Pathogens are

Field Sampling and Study Area
Four-hundred and sixty-nine rock partridges were collected in the French Alps from 1987 to 2019 (no samples collected during the period [1997][1998][1999]. For more details about the sampled animals please refer to Supplementary materials-Tables S1 and S2. Most of the animals derived from birds killed during hunting activity, with a few others found dead for other reasons (impact with cables, or predation). All the samples were geo-referenced, with the exception of eight animals for which it was not possible to obtain accurate spatial information (Supplementary materials- Figure S1).

Parasitological Analysis
The gizzard of each partridge was examined following the common parasitological standard techniques [38]. Adult worms were counted and collected under a stereoscopic microscope, and identified using a light microscope. The morphological identification of the nematodes was done according to the keys described by Skrjabin et al. [39] and Ebrahimi et al. [40].

Epidemiological Descriptors
Epidemiological indexes of prevalence (number of infected animals/total animals) and intensity (number of parasites/positive animals) were computed for the whole population, by sex and age. Afterwards, prevalence data were compared using Fisher's exact test, and frequency distribution of parasite intensity and abundance using the Kruskal-Wallis test [41]. To evaluate and represent the temporal trend of infection, a five-year aggregate prevalence was calculated and plotted. Data were aggregated at five-year intervals to minimize the unbalanced distribution of samples among years, and to derive a more robust trend.

Environmental Variables
The variables used to build the models are reported in Table 1. Spatial dataset layers include bio-climatic [42] and topographical factors, represented as raster data. All the rasters were clipped using the French Alps shapefile as a mask layer [43], considered as a proxy for the host species' potential distribution. The rasters were rescaled at a resolution of 1 km 2 , aligned, and re-projected using the same coordinate reference system (CRS-WGS 84/UTM zone 32N). All spatial data manipulation and representation were done using QGIS 3.6 [44].

Modelling Approach
The modeling approach was set in two steps with the following objectives: (i) identify and interpret the relationships between parasite occurrence (parasite presence) and environmental covariates, and (ii) build a spatial model to predict the parasite habitat suitability across the French Alps (39,372 km 2 ). In the first step, a tree classification model was built with the CARET package [45] in the R statistical framework [46]. Data was split into training (70% of points) and testing (30% of points) datasets. Resampling consisted in 10-fold cross-validated repeated three times and information gain was used as a splitting criterion. Although basic decision tree models are straightforward to interpret, they poorly predict the dependent variable for new values of the covariate. This limitation is due to the fact that they provide only a single description on the relationships between response and predictors [30]. Thus, in the second step, we built a random forest model (as implemented in the R package "randomForest" [47] to make more robust predictions. The following parameters were used to define the random model: (i) type of random forest = regression, (ii) number of trees = 500, (iii) number of variables tried at each split = 4. Variable importance was assessed based on mean decrease in accuracy, and the area under the curve value (AUC) was computed to evaluate model performances. AUC compares the model sensitivity (true positives) against "1-specificity" (false positives) over the entire range of the threshold. An excellent model has an AUC close to 1, whereas a poor model has an AUC close to 0 [48]. Model outputs were used to map the probability of occurrence of C. hamulosa. This first output was then converted into a binary surface using a threshold value maximizing both sensitivity and specificity. The percentages of suitable and unsuitable areas were computed, and a final map was created in QGIS v 3.6 [44]. Model outputs were reviewed and validated by wild Galliformes experts.
The 5-year aggregated prevalence trend is reported in Figure 1. Prevalence was almost duplicate in the period 1992-1996 to 2005-2009, when the highest prevalence (P: 67%, CI 95%: 54-80) was detected. This was followed by a 70% reduction in the most recent period of study ( Figure 1). Detailed information about (i) prevalence with 95% CI, and (ii) mean parasite intensity with standard deviation during the period, is available in the Supplementary materials- Table S3. The tree classification model indicates that few variables could discriminate parasite presence. Latitude was the most important predictor, with higher parasite presence probability at lower latitude. At sites with higher latitude, temperature seasonality (bio4) and altitude (dem) were important in further characterizing the distribution. Annual mean temperature (bio1), the amount of precipitation of the coldest quarter of the year (bio19), and altitude (dem) played an important role at lower latitude ( Figure 2). In agreement with the classification tree, the random forest model identified latitude as the most important factor. Altitude (dem), temperature seasonality (bio4), and the amount of precipitation of coldest quarter (bio19) were also important factors selected in this model framework ( Figure 3). Based on the results of the models, the probability of C. hamulosa occurrence was plotted, showing a decreasing trend, with a south-to-north and west-to-east gradient. The area with the lowest probability of parasite occurrence was located in the central part of the study area, at the border with the Italian Alps ( Figure 4).  Figure 5, the application of a cut-off discriminant value depicted the area considered suitable for parasite occurrence. The total area suitable for the parasite development consists of 73% (28,797 km 2 /39,371 km 2 ) of the French Alps, mostly located in the southern part of the study area. As regards model accuracy, both models performed quite well (decision tree AUC: 0.66, random forest AUC: 0.79).

Discussion
This study presents the first epidemiological data available on C. hamulosa infection in the rock partridge. Both prevalence (P: 39%), mean abundance (3), and mean intensity (7.7) values were quite high, with a peak in prevalence for the period 2005-2009. Our models depicted a high probability of C. hamulosa occurrence in the southern part of the French Alps, identified as the areas with the most suitable conditions (high probability of occurrence) for the parasite. Both models in our analysis highlighted the importance of latitude in explaining the observed parasite distribution patterns. Secondary, but still relevant, factors were altitude and bioclimatic variables (bio4, bio1, bio19-see Table 1).
Rock partridge (A. g. saxatilis) is an important gamebird, and at the same time a species with significant conservation value, given its recent global population decline [4]. However, information about the sanitary status of this Galliform in the Alps, including its parasite community, is still scarce. Among the available studies it is worthwhile citing the following surveys: Rizzoli et al. [49] in Trentino (Eastern Alps) in the 1990s, the most recent works of Formenti et al. [50] and Fanelli et al. [51] in the Western Italian Alps. A few other works have described the parasite community of the rock partridge [52], but C. hamulousa has never been reported on the Italian side of the Alps.
This nematode is found in the gizzards of different families of birds, mainly in the region of the caudal blind sac [25]. Comparable to our study, a similar C. hamulosa prevalence (P: 30%) was found by Ebrahimi et al. [40] in wild partridges (Alectoris chukar) in Iran; whereas studies on domestic birds reported a lower prevalence: 3.6% in chickens (Gallus g. domesticus L., 1758) from India [26], 4% in fowls from Iran [53], and 14.3% in backyard pheasants and 26.7% in backyard chickens from Brazil [54]. The large variation of prevalence values is mainly related to the origin of the samples, with the risk of infection being higher in wild Galliformes than poultry [55].
In the host population, parameters of occurrence of parasitic nematodes may vary significantly in relation to their age and sex. Generally, males experience a greater degree of parasitism than females, although these differences can be small [56,57]. Results from this study demonstrate that this is not necessarily true, as the C. hamulosa load was significantly higher in females than in males. This study also clarified the age-prevalence relationship of the parasite, which increased with age, the prevalence being significantly higher in adults. It is difficult to explain this result considering that the diet of young and adults is similar.
The trend of the five-year aggregated data shows a large variation in the prevalence along the study period. Even if our data do not support any cause-effect hypotheses, this finding might be linked to several factors, including a change in the rock partridge population density. The spread of parasites at population level is usually related to host abundance, with a direct relationship between the size of host population and parasite prevalence [58]. This is in line with the data reported by the Mountain Galliformes Observatory, who described a sharp increase in the rock partridge numbers in the French Alps during the period 2003-2007, followed by a population reduction (Observatoire des Galliformes de Montagne-http://www.observatoire-galliformes-montagne.com/, accessed on June 2020). From this perspective, even if more robust data should be obtained, this observation might indicate an effect of C. hamulosa in rock partridge population dynamics. Rock partridge is a species with demonstrated cyclic fluctuations [1], and the factors driving the cycles are still to be further studied. In other Galliformes, Trichostrongylus tenuis has been demonstrated to cause cyclic fluctuations of red grouse Lagopus lagopus scoticus [59]. Even if this hypothesis is yet to be demonstrated, it is possible that something similar occurs in rock partridges. The work of Rizzoli et al. [49] has in fact demonstrated that parasites may reduce partridge fecundity, and that the intensity of parasites is greater in cyclic than the noncyclic populations [10].
Another possible factor explaining the observed trend for large variation of the prevalence through the study period might be related to climatic factors. In fact, temperature and humidity have been recognized as major drivers of parasite presence and distribution [60]. Our work also provides insights into the spatial patterns of C. hamulosa. According to our estimates, about 73% of the French Alps is suitable for the development of the parasitic cycle. Latitude seems to be the environmental factor with the highest contribution to shaping the parasite distribution. Interestingly, another study carried out in the French Alps has recently associated the occurrence of helminths in the rock ptarmigan with latitude, interpreting this finding as a proxy of climatic and environmental conditions required for the development of free-living stages of the parasite. In contrast with our results, this work found an opposite gradient with the Northern part of the French Alps considered at higher risk of infection. However, the results in this case were related to a different host species (Lagopus muta) and a different parasite (Capillaria caudinflata) [29]. Indeed, it is important to highlight that parasite species distribution is largely determined by their hosts, and the variation in host abundance may influence parasites' geographical range [61]. Additionally, C. hamulosa occurrence is strictly influenced by the density and activity of the arthropod intermediate hosts [26], with complex interactions among environmental and climatic factors. Unfortunately, this information (definitive and intermediate host distribution and density), which could be valuable to improve the performance of our predictions, is largely incomplete or missing.
To take into account host species, parasites, intermediate hosts, and environmental interactions, we used machine learning algorithms to better fit the complexity of ecological systems [62]. Model results should be interpreted on the base of the non-linear associations between the response (parasite occurrence) and the explanatory variables. In this sense, the occurrence of C. hamulosa at identified high altitude cut-off values might reflect the distribution of the intermediate hosts, which are mostly found in the high mountain pastures [63].
Remarkably, in all the splits of the model the presence of the parasite was associated with higher values of annual mean temperature (bio1) and temperature seasonality (bio4), whereas high amounts of precipitation in the coldest quarter negatively (bio19) affected its presence. The precipitations in this case have to be considered as the amount of snow, and its permanence on the soil. These results corroborate the importance of climatic variables on the distribution of parasites with arthropods as intermediate host. In fact, seasonal climatic variation has been shown to influence the activity and abundance of the intermediate hosts, while lower temperature in winter has been claimed to induce arrested development of parasites in both the intermediate and final hosts [26].
To our knowledge, this is the first study investigating the infection of C. hamulosa in the rock partridge. Based on our results, we concluded that this parasite shows a large distribution in the sampled animals, and a potential wide distribution in the French Alps, based on the results of our model. Considering the potential impact of the parasite on population dynamics, the presence of C. hamulosa in the rock partridge in the French Alps must be taken into account when planning conservation strategies of this threatened species. Considering the scarce literature available on C. hamulosa, our work highlights the need for close sanitary monitoring in the areas with a higher risk of infection.

Conclusions
It is worth noting that data on hunting wildlife provide valuable information on diseases circulating among the animal's populations. However, studies using hunted animals might be subjected to bias due to the fact that disease also drastically increases the risk of being hunted. Despite this, we believe that such bias is minimal in our study, and that the birds sampled in this study did not differ much from the bulk of the population. This is due to the fact that this work covered a wide part of the species range in France, and was characterized by a long-term sampling (1987 to 2019).

Supplementary Materials:
The following are available online at http://www.mdpi.com/1424-2818/12/12/484/s1, Figure S1: Map showing the locations where the birds were sampled, Table S1: Number of birds sampled per year from 1987 to 2019, Table S2: Sample characteristics: age and sex of the birds sampled, Table S3